Mines ☛ Rivers ☛ Yields

Downstream Mining Impacts on Agriculture in Africa

Lukas Vashold (lvashold@wu.ac.at)
Gustav Pirich (gustav@pirich.at)
Max Heinze (mheinze@wu.ac.at)
Nikolas Kuschnig (nkuschnig@wu.ac.at)

Department of Economics, Vienna University of Economics and Business (WU Vienna)

May 16, 2025

 

 

 

Snapshot

Background

Methods & Data

Results

Mines — Curse or Blessing?

A Blessing?

A Curse?

How Pollution Travels

If water pollution from mines affects vegetation, we should observe reduced vegetation health downstream of a mine.

Using a remotely-sensed vegetation index, we find evidence for this.

 

 

Snapshot

Background

Methods & Data

Results

Appendix

Mining and Water Pollution

2015

2025

Research Question

What is the causal effect of water pollution from mining on agricultural productivity in Africa?

 

Snapshot

Background

Methods & Data

Results

Appendix

 

How to Find Affected Areas

Our unit of observation is the river basin. Lehner & Grill (2013) provide a nested basin collection, of which we use the most granular level.

If we spill a cup of water anywhere in a basin, it always ends up in the next basin downstream.

Water moves from upstream to downstream of a mine.

show actual basins

Intuition

Mines

We use mine locations from Maus et al. (2022), which includes some ASM sites.

We then designate mine basins and determine 10 levels each of upstream and downstream basins.

Variables and Observations

Outcome

  • We use the Enhanced Vegetation Index (EVI), which
    • is remotely sensed, and
    • ranges between –1 (water) and 1 (dense vegetation).

annual max. on all areas covered by vegetation

annual max. on areas covered by croplands

🠖 Max. EVI

🠖 Max. Cropl. EVI

Observations and Covariates

6,307 upstream basins

1,900
mine basins

6,127 downstream basins

Empirical Strategy (Spatial RDD)

We estimate:

\[ y_{imt} = \colorbox{var(--secondary-color-lightened)}{$\phantom{_i}\boldsymbol{\beta}'$}\:\colorbox{var(--primary-color-lightened)}{$F(x_i)\phantom{'}$}+\boldsymbol{\theta}'W_{it}+\mu_m+\psi_t+\varepsilon_{imt}, \]

where we let \(\colorbox{var(--primary-color-lightened)}{\(F(x_i)\)}\) return indicators:

\[ \colorbox{var(--primary-color-lightened)}{$f(x_i)_j\phantom{'}$} = \mathbb{I}(x=j)\qquad\text{for } j \in\{-10,\dots,-2,0,1,2,\dots,10\}. \]

  • \(y_{imt}\): outcome of basin \(i\) near mine \(m\) in year \(t\),
  • \(\mu_m\) and \(\psi_t\): mine and year fixed effects,
  • \(\boldsymbol{W}_{it}\): basin-specific covariates.
  • Parameter \(\colorbox{var(--secondary-color-lightened)}{\(\boldsymbol{\beta}'\)}\) is identified under the assumption that there are no other discontinuous changes at the mine basin.

Identification Assumption

  • To validate our results, we conduct a battery of robustness checks to address potential threats to identification.
    1. Extensive set of controls: geophysical features (elevation, slope, distance to coast, soil composition), meteorological conditions (temperature, precipitation), and socioeconomic indicators (population, accessibility, conflict)
    2. placebo outcomes show more
    3. matching procedure show more
    4. permutation test show more

Snapshot

Background

Methods & Data

Results

Appendix

 

 

Results Overview

  • We find a significant reduction in vegetation health downstream of mines.
  • Impacts are particularly strong in fertile regions and where gold mining predominates.
  • These results are robust to varying the sample, the outcome measurement, and the level of fixed effects.

Results

Peak vegetation is reduced by 1.28–1.35% relative to the mean on an affected area of 255,000 km².

For croplands, the estimates imply a 1.38–1.47% reduction across 74,000 km².

Outcome Peak Vegetation Peak Cropland Veg.
(Plain) (Full) (Plain) (Full)
Pooled Order
Downstream (1st–3rd) -0.0057*** -0.0056*** -0.0064** -0.0068***
(0.0018) (0.0020) (0.0025) (0.0026)
Fit statistics
  Observations 110,576 110,524 93,036 93,000
  R2 0.903 0.908 0.816 0.822
Controls
  Geophysical No Yes No Yes
  Meteorological No Yes No Yes
  Socioeconomic No Yes No Yes
Fixed-effects
  Year (2016–2023) Yes Yes Yes Yes
  Mine Yes Yes Yes Yes
Clustered (by mine-basin) standard errors in parentheses.
Significance levels: ***: 0.01, **: 0.05, *: 0.1.

Results – What Do These Impacts Mean?

  • We use survey data from farmers in Africa from IFPRI (2020) to estimate the crop yield–EVI elasticity.
  • Our measure is highly predictive of crop yields. show more
  • We estimate a 2.16–2.31% decrease in the value of overall crop production.
  • This amounts to a reduction in agricultural production of about 91,000 metric tons of cereals,
  • comparable to 5.4% of the 1.7 million tons that the World Food Program WFP distributes annually.

How Far Downstream Do These Effects Persist?

Heterogeneity – Which Areas Are Affected?

Heterogeneity – Which Areas Are Affected?

  • Stronger effects in West Africa and on highly fertile croplands.
  • Gold mining regions show 75% larger impacts on croplands.

Robustness

Impact Decay Assessment

We re-estimate the main specification using an exponential distance decay function, \(\mathrm{exp}(-\delta d_{ij})\), where \(d_{ij}\) is the distance along the river from a mine.

More on Water Pollution

  • We identify the impact of water pollution from mining on agricutural productivity.
  • Water quality data is sparse, and limited to South Africa.
  • We provide suggestive evidence of levels of water pollution as a mediating mechanism (United Nations Environment Programme, 2025)
  • We find elevated levels of key pollutant measures for mine and downstream basins. show more

Limitations

  • Remotely sensed vegetation indices capture crop yields only indirectly.
  • Differentiating between artisanal and industrial mining remains challenging due to data scarcity and co-location of mines.
  • No direct evidence on farmers’ adaptive responses (e.g., migration, altered crop choices, or barrier to irrigation).
  • Noise from multiple data sources likely leads to attenuation of true estimates.

Conclusion

We identified the causal effects of mining

  • on agricultural productivity,
  • mediated by water pollution.

Our results showed a negative impact on vegetation health.

Effects were particularly strong for larger mines, gold mining regions, and in regions with highly fertile croplands.

Results were robust to changes of treatment, outcome, sample, methods, and estimation procedures.

References

This list is scrollable.

Abatzoglou, J. T., Dobrowski, S. Z., Parks, S. A., & Hegewisch, K. C. (2018). TerraClimate, a high-resolution global dataset of monthly climate and climatic water balance from 19582015. Scientific Data, 5(170191), 1–12. https://doi.org/10.1038/sdata.2017.191
Amatulli, G., Domisch, S., Tuanmu, M.-N., Parmentier, B., Ranipeta, A., Malczyk, J., & Jetz, W. (2018). A suite of global, cross-scale topographic variables for environmental and biodiversity modeling. Scientific Data, 5(180040), 1–15. https://doi.org/10.1038/sdata.2018.40
ASM Inventory. (2022). World maps of artisanal and small-scale mining. http://artisanalmining.org/Inventory/
Azzari, G., Jain, M., & Lobell, D. B. (2017). Towards fine resolution global maps of crop yields: Testing multiple methods and satellites in three countries. Remote Sensing of Environment, 202, 129--141. https://doi.org/10.1016/j.rse.2017.04.014
Bazillier, R., & Girard, V. (2020). The gold digger and the machine. Evidence on the distributive effect of the artisanal and industrial gold rushes in Burkina Faso. Journal of Development Economics, 143, 102411. https://doi.org/10.1016/j.jdeveco.2019.102411
Becker-Reshef, I., Vermote, E., Lindeman, M., & Justice, C. (2010). A generalized regression-based model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data. Remote Sensing of Environment, 114(6), 1312--1323. https://doi.org/10.1016/j.rse.2010.01.010
Bolton, D. K., & Friedl, M. A. (2013). Forecasting crop yield using remotely sensed vegetation indices and crop phenology metrics. Agricultural and Forest Meteorology, 173, 74--84. https://doi.org/10.1016/j.agrformet.2013.01.007
Cattaneo, M. D., Idrobo, N., & Titiunik, R. (2019). A practical introduction to regression discontinuity designs: foundations. Cambridge University Press. https://doi.org/10.1017/9781108684606
Food and Agriculture Organization of the United Nations, International Fund of Agricultural Developmenmt, United Nations International Children’s Emergency Fund, United Nations World Food Programme, & World Health Organization. (2023). The state of food security and nutrition in the world 2023. Urbanization, agrifood systems transformation and healthy diets across the rural–urban continuum [Report]. https://doi.org/10.4060/cc3017en
Girard, V., Molina-Millán, T., Vic, G., et al. (2022). Artisanal mining in africa. In Working paper series no. 2201. Novafrica.
Goltz, J. von der, & Barnwal, P. (2019). Mines: The local wealth and health effects of mineral mining in developing countries. Journal of Development Economics, 139, 1–16. https://doi.org/10.1016/j.jdeveco.2018.05.005
Hengl, T., Jesus, J. M. de, Heuvelink, G. B. M., Gonzalez, M. R., Kilibarda, M., Blagotić, A., Shangguan, W., Wright, M. N., Geng, X., Bauer-Marschallinger, B., Guevara, M. A., Vargas, R., MacMillan, R. A., Batjes, N. H., Leenaars, J. G. B., Ribeiro, E., Wheeler, I., Mantel, S., & Kempen, B. (2017). SoilGrids250m: Global gridded soil information based on machine learning. PLOS ONE, 12(2), e0169748. https://doi.org/10.1371/journal.pone.0169748
Hund, K., La Porta, D., Fabregas, T. P., Laing, T., & Drexhage, J. (2023). Minerals for climate action: The mineral intensity of the clean energy transition. World Bank Group. https://pubdocs.worldbank.org/en/961711588875536384/Minerals-for-Climate-Action-The-Mineral-Intensity-of-the-Clean-Energy-Transition.pdf
IFPRI. (2020). AReNA’s DHS-GIS database (Version V1) [Dataset]. International Food Policy Research Institute (IFPRI). https://doi.org/10.7910/DVN/OQIPRW
International Council on Mining and Metals. (2022). Mining contribution index (MCI) (6th Edition). https://www.icmm.com/website/publications/pdfs/social-performance/2022/research_mci-6-ed.pdf?cb=16134
Jasansky, S., Lieber, M., Giljum, S., & Maus, V. (2023). An open database on global coal and metal mine production. Scientific Data, 10(52), 1–12. https://doi.org/10.1038/s41597-023-01965-y
Johnson, D. M. (2016). A comprehensive assessment of the correlations between field crop yields and commonly used MODIS products. International Journal of Applied Earth Observation and Geoinformation, 52, 65--81. https://doi.org/10.1016/j.jag.2016.05.010
Kolesár, M., & Rothe, C. (2018). Inference in regression discontinuity designs with a discrete running variable. American Economic Review, 108(8), 2277–2304. https://doi.org/10.1257/aer.20160945
Kossoff, D., Dubbin, W. E., Alfredsson, M., Edwards, S. J., Macklin, M. G., & Hudson-Edwards, K. A. (2014). Mine tailings dams: Characteristics, failure, environmental impacts, and remediation. Applied Geochemistry, 51, 229–245. https://doi.org/10.1016/j.apgeochem.2014.09.010
Lehner, B., & Grill, G. (2013). Global river hydrography and network routing: Baseline data and new approaches to study the world’s large river systems. Hydrological Processes, 27(15), 2171--2186. https://doi.org/10.1002/hyp.9740
Macklin, M. G., Thomas, C. J., Mudbhatkal, A., Brewer, P. A., Hudson-Edwards, K. A., Lewin, J., Scussolini, P., Eilander, D., Lechner, A., Owen, J., Bird, G., Kemp, D., & Mangalaa, K. R. (2023). Impacts of metal mining on river systems: A global assessment. Science, 381(6664), 1345–1350. https://doi.org/10.1126/science.adg6704
Maus, V., Giljum, S., Silva, D. M. da, Gutschlhofer, J., Rosa, R. P. da, Luckeneder, S., Gass, S. L., Lieber, M., & McCallum, I. (2022). An update on global mining land use. Scientific Data, 9(1), 1–11.
Moura, A., Lutter, S., Siefert, C. A. C., Netto, N. D., Nascimento, J. A. S., & Castro, F. (2022). Estimating water input in the mining industry in Brazil: A methodological proposal in a data-scarce context. Extractive Industries and Society, 9, 101015. https://doi.org/10.1016/j.exis.2021.101015
Padilla, A. J., Otarod, D., Deloach-Overton, S. W., Kemna, R. F., Freeman, P. A., Wolfe, E. R., Bird, L. R., Gulley, A. L., Trippi, M. H., Dicken, C. L., Hammarstrom, J. M., & Brioche, A. S. (2021). Compilation of geospatial data (GIS) for the mineral industries and related infrastructure of Africa [Data release]. U.S. Geological Survey. https://doi.org/10.5066/P97EQWXP
Ruppen, D., Runnalls, J., Tshimanga, R. M., Wehrli, B., & Odermatt, D. (2023). Optical remote sensing of large-scale water pollution in Angola and DR Congo caused by the Catoca mine tailings spill. International Journal of Applied Earth Observation and Geoinformation, 118, 103237. https://doi.org/10.1016/j.jag.2023.103237
Schwarzenbach, R. P., Egli, T., Hofstetter, T. B., Gunten, U. von, & Wehrli, B. (2010). Global water pollution and human health. Annual Review of Environment and Resources, Volume 35, 2010, 109–136. https://doi.org/10.1146/annurev-environ-100809-125342
United Nations Environment Programme. (2025). GEMStat database of the global environment monitoring system for freshwater (GEMS/water) Programme. International Centre for Water Resources; Global Change, Koblenz.
Weiss, D. J., Nelson, A., Gibson, H. S., Temperley, W., Peedell, S., Lieber, A., Hancher, M., Poyart, E., Belchior, S., Fullman, N., Mappin, B., Dalrymple, U., Rozier, J., Lucas, T. C. D., Howes, R. E., Tusting, L. S., Kang, S. Y., Cameron, E., Bisanzio, D., … Gething, P. W. (2018). A global map of travel time to cities to assess inequalities in accessibility in 2015. Nature, 553, 333–336. https://doi.org/10.1038/nature25181

Background

Methods & Data

Results

Appendix

 

 

 

Appendix · Basins

Illustration from Lehner & Grill (2013)

go back

Appendix · A Proxy for Agricultural Activity

We get a proxy for agricultural productivity like this:

  1. Filter out cloud cover.
  2. Aggregate the mean EVI per basin.
  3. Take the annual maximum per basin per year. 🠖 Max. EVI
  4. Apply a cropland mask. 🠖 Max. Cropland EVI

This peak vegetation index has been shown to proxy well for crop yields (Azzari et al., 2017; Becker-Reshef et al., 2010; Bolton & Friedl, 2013; Johnson, 2016).

Appendix · Summary Statistics

Variable Unit NT Mean St. Dev. Min. Max.
Peak Vegetation Index [−1, 1] 110,576 0.428 0.154 0.016 0.993
Mean Vegetation Index [−1, 1] 110,576 0.279 0.112 −0.021 0.578
Peak Cropland Veg. Index [−1, 1] 93,036 0.464 0.133 −0.068 0.978
Mean Cropland Veg. Index [−1, 1] 93,036 0.298 0.101 −0.104 0.601
Elevation Meters 110,568 820.4 481.1 −118.3 3,059.7
Slope Degrees 110,568 2.23 2.34 0.0 20.9
Max. Temperature Degree Celsius 110,572 34.3 3.9 15.6 48.8
Precipitation Millimeters 110,576 901.8 595.2 0.64 4,456.7
Population Capita 110,576 8,471 37,716 0.0 1,396,921
Accessibility Minutes 110,528 164.3 179.1 1.0 2,659.9

Appendix · Impact Decay Assessment

  • We re-estimate our main specification using an exponential decay function: \(\mathrm{exp}(-\delta d_{ij})\).
  • Hydrological studies on dispersion patterns suggest using an exponential decay function.
  • Since the decay parameter is unknown, we conduct a grid search for \(\delta\in[0.001,2]\).
  • We then use a Bayesian Model Averaging approach with BIC as marginal likelihood approximation.
  • Finally, we compute the mean effect decay at increasing distances.

Appendix · Basin Numbers

The number of mine-basins with Y upstream and X downstream basins in the dataset. go back

Appendix · Basins by Order

Order Upstream Downstream
N Distance N Distance
0 (1900) (0.0) (1900) (0.0)
1 847 13.9 1162 11.1
2 781 24.5 882 22.0
3 722 35.0 743 32.7
4 698 44.9 643 43.3
5 653 55.3 578 54.0
6 576 66.3 512 64.3
7 562 75.8 458 74.1
8 522 86.5 416 84.4
9 494 95.8 382 95.0
10 452 104.2 351 104.7

Appendix · Full Order Specifiaction Results

Outcome Peak Vegetation Peak Cropland Veg.
(Specification) (Plain) (Full) (Plain) (Full)
Individual Order
Downstream (1st) -0.0045*** -0.0043** -0.0051** -0.0050**
(0.0017) (0.0018) (0.0025) (0.0025)
Downstream (2nd) -0.0049** -0.0048** -0.0058* -0.0067**
(0.0022) (0.0024) (0.0031) (0.0032)
Downstream (3rd) -0.0085*** -0.0087*** -0.0088** -0.0099***
(0.0028) (0.0029) (0.0037) (0.0038)
Downstream (4th) -0.0049* -0.0062* −0.0029 −0.0044
(0.0030) (0.0033) (0.0038) (0.0040)
Downstream (5th) −0.0034 −0.0053 0.0007 −0.0016
(0.0033) (0.0037) (0.0042) (0.0045)
Fit statistics
  Observations 110,576 110,524 93,036 93,000
  R2 0.903 0.908 0.816 0.822
Pooled Order
Downstream (1st–3rd) -0.0057*** -0.0056*** -0.0064** -0.0068***
(0.0018) (0.0020) (0.0025) (0.0026)
Fit statistics
  Observations 110,576 110,524 93,036 93,000
  R2 0.903 0.908 0.816 0.822
Controls
  Geophysical No Yes No Yes
  Meteorological No Yes No Yes
  Socioeconomic No Yes No Yes
Fixed-effects
  Year (2016–2023) Yes Yes Yes Yes
  Mine Yes Yes Yes Yes
Clustered (Mine) standard-errors in parentheses
Signif. Codes: ***: 0.01, **: 0.05, *: 0.1

Appendix · Distance Specification Results

Dependent Variables: Maximum Vegetation EVI Maximum Croplands EVI
Model: (1) (2) (3) (4) (5) (6) (7) (8)
Linear distance
Downstream −0.0050** −0.0045** −0.0033 −0.0034 −0.0050* −0.0049* −0.0041 −0.0042
(0.0023) (0.0022) (0.0022) (0.0022) (0.0029) (0.0029) (0.0029) (0.0029)
Downstream × Distance −7.57×10−6 −3.59×10−5 −8.32×10−5 −8.47×10−5 1.47×10−5 −4.19×10−6 −5.85×10−5 −5.96×10−5
(4.69×10−5) (5.36×10−5) (5.38×10−5) (5.32×10−5) (5.85×10−5) (6.91×10−5) (6.96×10−5) (6.94×10−5)
Distance 7.75×10−6 3.26×10−5 5.61×10−5 6.18×10−5 2.75×10−5 4.08×10−5 6.32×10−5 5.66×10−5
(3.91×10−5) (4.13×10−5) (4.12×10−5) (4.04×10−5) (4.97×10−5) (5.45×10−5) (5.45×10−5) (5.3×10−5)
Fit statistics
Observations 110,576 110,568 110,564 110,524 93,036 93,036 93,032 93,000
R2 0.90282 0.90452 0.90762 0.90783 0.81609 0.81748 0.82138 0.82165
Linear–quadratic distance
Downstream −0.0056** −0.0055** −0.0050** −0.0052** −0.0077** −0.0076** −0.0072** −0.0073**
(0.0027) (0.0026) (0.0025) (0.0025) (0.0035) (0.0036) (0.0035) (0.0035)
Downstream × Distance 2.64×10−5 2.01×10−5 5.75×10−6 5.45×10−6 0.0002 0.0001 0.0001 0.0001
(0.0001) (0.0001) (0.0001) (0.0001) (0.0002) (0.0002) (0.0002) (0.0002)
Downstream × Distance2 −3.04×10−7 −4.7×10−7 −7.27×10−7 −7.35×10−7 −1.2×10−6 −1.17×10−6 −1.38×10−6 −1.36×10−6
(8.52×10−7) (8×10−7) (8.09×10−7) (7.99×10−7) (1.2×10−6) (1.2×10−6) (1.18×10−6) (1.18×10−6)
Distance 3.97×10−5 3.93×10−5 3.33×10−5 3.64×10−5 −4.23×10−6 1.24×10−6 1.17×10−5 −1.21×10−6
(9.08×10−5) (8.61×10−5) (8.93×10−5) (8.63×10−5) (0.0001) (0.0001) (0.0001) (0.0001)
Distance2 −2.43×10−7 −5.04×10−8 1.76×10−7 1.97×10−7 2.55×10−7 3.18×10−7 4.13×10−7 4.64×10−7
(6.32×10−7) (5.85×10−7) (6.05×10−7) (5.91×10−7) (9.26×10−7) (9.37×10−7) (9.11×10−7) (9.2×10−7)
Fit statistics
Observations 110,576 110,568 110,564 110,524 93,036 93,036 93,032 93,000
R2 0.90283 0.90453 0.90762 0.90784 0.81612 0.81751 0.82142 0.82168
Exponential decay
δ = 0.005 δ = 0.006 δ = 0.002 δ = 0.002 δ = 0.035 δ = 0.035 δ = 0.020 δ = 0.010
exp{−δ×Distance} × Downstream −0.0062*** −0.0062*** −0.0060*** −0.0062*** −0.0093*** −0.0091*** −0.0074** −0.0068**
(0.0023) (0.0023) (0.0023) (0.0023) (0.0034) (0.0033) (0.0029) (0.0029)
Fit statistics
Observations 110,576 110,568 110,564 110,524 93,036 93,036 93,032 93,000
R2 0.901147 0.902842 0.905958 0.906169 0.812592 0.813949 0.817862 0.818141
Fixed-effects
Year Yes Yes Yes Yes Yes Yes Yes Yes
Mine Yes Yes Yes Yes Yes Yes Yes Yes
Clustered (mine basin) standard-errors in parentheses    Significance: ***: 0.01, **: 0.05, *: 0.1

Appendix · Varying Sample Definition

Dependent Variables: Maximum EVI Maximum Cropland EVI
Model: (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Variables
Downstream × Order = 0 −0.0059*** −0.0076*** −0.0062***     −0.0095*** −0.0082*** −0.0094***    
(0.0013) (0.0014) (0.0012)     (0.0020) (0.0024) (0.0022)    
Downstream × Order = 1 −0.0057*** −0.0053*** −0.0053*** −0.0049** −0.0051** −0.0061** −0.0049 −0.0051* −0.0061** −0.0069*
(0.0017) (0.0020) (0.0017) (0.0020) (0.0021) (0.0026) (0.0032) (0.0030) (0.0030) (0.0039)
Downstream × Order = 2 −0.0066*** −0.0054**   −0.0056**   −0.0062** −0.0057   −0.0062*  
(0.0021) (0.0026)   (0.0023)   (0.0030) (0.0037)   (0.0033)  
Fixed-effects
Year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Mine Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fit statistics
Observations 114,496 61,712 32,360 99,320 9,168 94,604 50,914 27,589 81,278 7,623
R2 0.92395 0.91566 0.93993 0.92392 0.93378 0.78597 0.76613 0.84032 0.78332 0.81766
Within R2 0.05582 0.05702 0.05650 0.05511 0.07364 0.02531 0.02382 0.03446 0.02322 0.03884
Clustered (Mine) standard-errors in parentheses
Signif. Codes: ***: 0.01, **: 0.05, *: 0.1

Appendix · Research Design

Illustration of the research design. The comparison of up- and downstream basins enables the identification of mine impacts that are mediated by the river.

Appendix · Varying Outcome / Fixed Effects

Dependent Variables: Maximum EVI Mean EVI Maximum Cropland EVI Mean C EVI ESA C EVI
Model: (1) (2) (3) (4) (5) (6) (7) (8) (9)
Variables
Downstream × Order = 0 −0.0059*** −0.0065*** −0.0079*** −0.0048*** −0.0095*** −0.0104*** −0.0109*** −0.0073*** −0.0048*
(0.0013) (0.0013) (0.0014) (0.0009) (0.0020) (0.0020) (0.0021) (0.0013) (0.0026)
Downstream × Order = 1 −0.0057*** −0.0060*** −0.0066*** −0.0035*** −0.0061** −0.0062** −0.0064*** −0.0043** −0.0035
(0.0017) (0.0016) (0.0017) (0.0011) (0.0026) (0.0025) (0.0025) (0.0017) (0.0032)
Downstream × Order = 2 −0.0066*** −0.0064*** −0.0067*** −0.0038*** −0.0062** −0.0058** −0.0064** −0.0055*** −0.0015
(0.0021) (0.0020) (0.0020) (0.0013) (0.0030) (0.0029) (0.0028) (0.0019) (0.0035)
Fixed-effects
Year Yes Yes Yes Yes Yes Yes Yes Yes Yes
Mine Yes     Yes Yes     Yes Yes
Pfaffstetter basin level 8   Yes       Yes      
Pfaffstetter basin level 6     Yes       Yes    
Fit statistics
Observations 114,496 114,496 114,496 114,496 94,604 94,604 94,604 94,604 67,649
R2 0.92395 0.91954 0.90419 0.95707 0.78597 0.77061 0.74193 0.88641 0.80154
Within R2 0.05582 0.06500 0.08647 0.11783 0.02531 0.02957 0.04285 0.04478 0.02553
Clustered (Mine) standard-errors in parentheses
Signif. Codes: ***: 0.01, **: 0.05, *: 0.1

Appendix · Placebo Outcomes

Dependent Variables: Elevation Slope Max. Temp Precipitation Accessibility 2015 Population 2015
Model: (1) (2) (3) (4) (5) (6)
Variables
Downstream −6.852 −0.0538 −0.0137 0.6025 −5.427 2,125.7
(8.509) (0.0912) (0.0567) (3.934) (5.531) (1,589.8)
Distance × Downstream −5.008*** −0.0060 0.0135*** −0.1942 0.0839 −182.9***
(0.4814) (0.0044) (0.0036) (0.2860) (0.3278) (55.80)
Distance2 × Downstream 0.0043 −8.25×10−6 2.12×10−6 0.0003 0.0004 1.081***
(0.0039) (4.01×10−5) (3.36×10−5) (0.0020) (0.0028) (0.3463)
Distance 2.326*** 0.0025 −0.0067** 0.0879 0.7557*** −54.72
(0.4215) (0.0039) (0.0032) (0.2129) (0.2587) (45.17)
Distance2 0.0005 1.12×10−6 −5.34×10−6 −0.0005 −0.0013 0.3439
(0.0033) (3.49×10−5) (3.1×10−5) (0.0015) (0.0021) (0.2724)
Fixed-effects
Year Yes Yes Yes Yes Yes Yes
Mine Yes Yes Yes Yes Yes Yes
Fit statistics
Observations 114,616 114,616 114,616 114,616 114,576 114,536
R2 0.95627 0.70192 0.95579 0.96187 0.88768 0.59121
Within R2 0.41042 0.01108 0.07605 0.00070 0.04659 0.00851
Clustered (Mine) standard-errors in parentheses
Signif. Codes: ***: 0.01, **: 0.05, *: 0.1 go back to identification go back to results

Appendix · Placebo Outcomes

Order estimates when using elevation, slope, temperature, precipitation, accessibility to cities, and population as placebo outcomes. go back to identification go back to results

Appendix · Matching Exercise

Balance of elevation, slope, temperature, and precipitation before and after matching. (Soilgrids not pictured.) go back to identification go back to results

Appendix · Dist. Spec. w/ Aut. BW Selection (No Controls)

Max EVI Max C EVI
Conv. Bias-Corr. Conv. Bias-Corr.
No Controls
Conventional −0.0050*** −0.0056*** −0.0112*** −0.0116***
(0.0015) (0.0015) (0.0020) (0.0025)
Observations 37,880 37,880 32,813 32,813
Bandwidth (Conv) 20.3 20.3 20.7 20.7
Bandwidth (Bias) 46.4 46.4 47.4 47.4

Note: Table shows results for the estimation of the main specification, where distance in kilometres along the river network is the running variable. Bandwidths are chosen automatically following Cattaneo et al. (2019),using a triangular kernel, the mean-squared-error criterion, and bias correction. The upper-panel models include no covariates; the lower-panel models include the full set of controls. Columns (1) and (2) use overall EVI as the outcome, while columns (3) and (4) use cropland-specific EVI. Columns (1) and (3) fit a linear polynomial in distance on each side of the cutoff; columns (2) and (4) fit a quadratic polynomial. All specifications include mine and year fixed effects. Standard errors are clustered at the mine-basin system level.

Significance codes: *** p < 0.01, ** p < 0.05, * p < 0.1 · Clustered (Mine) standard errors in parentheses.

Appendix · Dist. Spec. w/ Aut. BW Selection (Full Controls)

Max EVI Max C EVI
Conv. Bias-Corr. Conv. Bias-Corr.
With Full Controls
Conventional −0.0045*** −0.0049*** −0.0100*** −0.0118***
(0.0015) (0.0015) (0.0020) (0.0026)
Observations 38,200 38,200 32,629 32,629
Bandwidth (Conv) 20.6 20.6 20.5 20.5
Bandwidth (Bias) 43.4 43.4 45.4 45.4

Note: Table shows results for the estimation of the main specification, with distance (in kilometres along the river network) as the running variable. Bandwidths are selected automatically following Cattaneo et al. (2019), using a triangular kernel, the mean-squared-error criterion, and bias correction. Models in the upper panel include no covariates; models in the lower panel include the full set of controls. Columns (1) and (2) report results for overall EVI, while columns (3) and (4) report results for cropland-specific EVI. Columns (1) and (3) fit a linear polynomial in distance on each side of the cutoff; columns (2) and (4) fit a quadratic polynomial. All specifications include mine and year fixed effects. Standard errors are clustered at the mine-basin system level.

Significance codes: *** p < 0.01, ** p < 0.05, * p < 0.1 · Clustered (Mine) standard errors in parentheses.

Appendix · Permutation – Robustness

Estimation results when the treatment status (i.e., whether basins are down- or upstream) is randomized (5,000 runs, balance between statuses is kept). The red crosses indicate estimation results for the main specification. go back to identification go back to results

Appendix · Ord. Spec. w/ Aut. BW Selection (Full Controls)

Max EVI Max C EVI
No Cluster Cluster (Mine Basin) No Cluster Cluster (Mine Basin)
No Controls
I(order > 0) −0.0048 −0.0048 −0.0090*** −0.0090**
(0.0013) (0.0019) (0.0018) (0.0030)
Observations 45,613 45,613 38,537 38,537
Bandwidth 2 2 2 2

Note: Table shows results for the estimation of the main specification, with distance defined by basin order relative to the mine basin as the running variable. Bandwidths are selected automatically following Kolesár & Rothe (2018), using a triangular kernel, the mean-squared-error criterion, and no bias correction. Models in the upper panel include no covariates; models in the lower panel include the full set of controls. Columns (1) and (2) report results for overall EVI, while columns (3) and (4) report results for cropland-specific EVI. Columns (1) and (3) use unclustered standard errors; columns (2) and (4) cluster standard errors at the mine-basin system level.
All specifications include mine and year fixed effects.

Significance codes: *** p < 0.01, ** p < 0.05, * p < 0.1

Appendix · Validation of Outcome IFPRI

Outcome: ln(Crops, Value) ln(Crops, FY) ln(Cereals, Value) ln(Cereals, Yield) ln(Cereals, FY)
Model: (1) (2) (3) (4) (5)
Variables
Max. Cropland EVI 3.398*** 0.9519*** 2.489*** 0.8995*** 0.5589**
(0.4230) (0.1828) (0.9150) (0.1586) (0.2704)
Fixed effects
Wave Yes Yes Yes Yes Yes
Fit statistics
Observations 44,682 44,380 44,682 44,682 44,171
R2 0.65336 0.35656 0.50120 0.60944 0.32956
Within R2 0.08225 0.00717 0.02177 0.02195 0.00153
Clustered (wave) standard-errors in parentheses  Significance: ***: 0.01, **: 0.05, *: 0.1

go back

Appendix · Water Pollution Data for South Africa

go back